[Gluster-users] poor dd performance w/ gluster 3.0.3

[Jeremy Enos]

What worked at 100MB/sec w/ gluster 3.0 is now 1.2MB/sec with 3.0.3.
I have 2 5 host stripes, replicated.  (I'm aware that this config is 
unsupported)  Just updated from 3.0 to 3.0.3.
While my IOR tests across 8 clients come out fine (slightly better than 
w/ 3.0) - dd performance has gone to crap.  While this is fine I guess 
as long as applications still perform well, I'm curious as to why dd is 
affected this way.  (it used to be ~100MB/sec w/ 3.0, now is 1.2MB/sec 
w/ 3.0.3)

     Jeremy

[jenos at ac31 IOR-2.10.1]$ ./runior.sh -t 1m -b 1g
IOR-2.10.1: MPI Coordinated Test of Parallel I/O

Run began: Thu Mar 18 17:58:02 2010
Command line used: ./IOR -a POSIX -q -wr -m -C -F -e -N 8 -t 1m -b 1g -o 
/scratch/jenos/iortest
Machine: Linux ac31.ncsa.uiuc.edu

Summary:
         api                = POSIX
         test filename      = /scratch/jenos/iortest
         access             = file-per-process
         ordering           = sequential offsets
         clients            = 8 (1 per node)
         repetitions        = 1
         xfersize           = 1 MiB
         blocksize          = 1 GiB
         aggregate filesize = 8 GiB

access    bw(MiB/s)  block(KiB) xfer(KiB)  open(s)    wr/rd(s)   
close(s)   iter
------    ---------  ---------- ---------  --------   --------   
--------   ----
write     180.90     1048576    1024.00    0.072115   45.25      
20.07      0
read      980.56     1048576    1024.00    0.032333   8.34       
3.56       0

Max Write: 180.90 MiB/sec (189.68 MB/sec)
Max Read:  980.56 MiB/sec (1028.19 MB/sec)

Run finished: Thu Mar 18 17:58:57 2010
[jenos at ac31 IOR-2.10.1]$ dd conv=fsync if=/dev/zero 
of=/scratch/jenos/bigfile bs=1024 count=1000000
^C489835+0 records in
489835+0 records out
501591040 bytes (502 MB) copied, 421.996 s, 1.2 MB/s